Data Systems Seminar Series (2016-2017)

We argue for a substantial shift away from human-crafted solutions and towards leveraging machine learning algorithms to address the above challenges. These algorithms can be trained on application- specific properties and customized performance goals to automatically learn how to provision resources as well as schedule the execution of incoming query workloads. Towards this vision, we have developed WiSeDB, a learning-based performance management service for cloud-deployed data management applications. In this talk, I will discuss how WiSeDB uses (a) supervised learning to automatically learn cost-effective models for guiding query placement, scheduling, and resource provisioning decisions for batch workload processing, and (b) reinforcement learning to naturally adapt to changes in query arrival rates and dynamic resource availability, while being decoupled from notoriously inaccurate performance prediction models.

Approximate lifted inference with probabilistic databases

Performing inference over large uncertain data sets is becoming a central data management problem. Recent large knowledge bases, such as Yago, Nell or DeepDive, have millions to billions of uncertain tuples. Because general reasoning under uncertainty is highly intractable, many state-of-the-art systems today perform approximate inference by reverting to sampling. This talk shows an alternative approach that allows approximate ranking answers to hard probabilistic queries in guaranteed polynomial time, and by using only basic operators of existing database management systems (i.e., no sampling required).

(1) The first part of this talk develops upper and lower bounds for the probability of Boolean functions by treating multiple occurrences of variables as independent and assigning them new individual probabilities. We call this approach dissociation and give an exact characterization of optimal oblivious bounds (i.e. when the new probabilities are chosen independent of the probabilities of all other variables). Our new bounds shed light on the connection between previous relaxation-based and model-based approximations and unify them as concrete choices in a larger design space.

(2) The second part then draws the connection to lifted inference and shows how the problem of approximate probabilistic inference can be entirely reduced to a standard query evaluation problem with aggregates. There are no iterations and no exponential blow-ups. All benefits of relational engines (such as cost-based optimizations, multi-core query processing, shared-nothing parallelization) are directly available to queries over probabilistic databases. To achieve this, we compute approximate rather than exact probabilities, with a one-sided guarantee: The probabilities are guaranteed to be upper bounds to the true probabilities, which we show is sufficient to rank the top query answers with high precision. We give experimental evidence on synthetic TPC-H data that this approach can be orders of magnitude faster and also more accurate than sampling-based approaches.

(Based on joint work with Dan Suciu from TODS 2014, VLDB 2015, and VLDBJ 2016: http://arxiv.org/pdf/1409.6052, http://arxiv.org/pdf/1412.1069, http://arxiv.org/pdf/1310.6257)

For several decades now, the amount of data available to us has been growing at a pace far higher than our ability to process it; this trend, popularly referred to as "big data", has accelerated many-fold in recent years with the emergence of efficient and mass-produced scientific instruments, increasing ease of generating and publishing data, and proliferation of Internet-connected devices. In this talk, I will present an overview of two recent projects from my group at UMD on building scalable platforms for large-scale data analytics.

First, I will discuss our ongoing work on building a platform, called "DataHub", for enabling collaborative data science, where teams of data scientists can simultaneously analyze, modify, and share datasets, to understand trends and to extract actionable insights. While numerous solutions exist for specific data analysis tasks, underlying infrastructure and data management capabilities for supporting ad hoc collaboration pipelines are still largely missing. I will present our vision for a unified, dataset-centric platform for addressing these challenges, and present our recent work on: (a) efficiently managing a large number versioned datasets, (b) designing and supporting a unified query language to seamlessly query versioning and provenance information, and (c) lifecycle management of complex machine learning models like deep neural networks.

Second, I will present our initial work on extracting hidden graphs from relational databases. Although there has been much work on large-scale graph analytics, graphs are not the primary representation choice for most data today, and users who want to employ graph analytics are forced to extract data from their data stores, construct the requisite graphs, and then use a specialized engine to write and execute their graph analysis tasks. I will describe our work on a system called GraphGen, that enables users to declaratively specify graph extraction tasks over relational databases, visually explore the extracted graphs, and write and execute graph algorithms over them, either directly or using existing graph libraries like the widely used NetworkX Python library.

A Hitchhiker's Guide to Ontology 

In this talk, I will give an overview of our recent work in the area of knowledge bases. I will first talk about our main project, the YAGO knowledge base. YAGO is now multilingual, and has grown into a larger project at the Max Planck Institute for Informatics and Télécom ParisTech. I will then talk about rule mining. We can find semantic correlations in the form of Horn rules in the knowledge base. In our newest work, we show how rule mining can be applied to predict the completeness or incompleteness of the data in the knowledge base. I will also talk about watermarking approaches to trace the provenance of ontological data. Finally, I will showcase our work on creativity in knowledge bases.

Data Profiling

Data profiling comprises a broad range of methods to efficiently analyze a given data set. In a typical scenario, which mirrors the capabilities of commercial data profiling tools, tables of a relational database are scanned to derive metadata, such as data types and value patterns, completeness and uniqueness of columns, keys and foreign keys, and occasionally functional dependencies and association rules. Individual research projects have proposed several additional profiling tasks, such as the discovery of inclusion dependencies or conditional functional dependencies.

Data profiling deserves a fresh look for two reasons: First, the area itself is neither established nor defined in any principled way, despite significant research activity on individual parts in the past. Second, current data profiling techniques hardly scale beyond what can only be called small data. Finally, more and more data beyond the traditional relational databases are being created and beg to be profiled. The talk highlights the state of the art and proposes new research directions and challenges.

The blooming of different cloud data management infrastructures, specialized for different kinds of data and tasks, has led to a wide diversification of DBMS interfaces and the loss of a common programming paradigm. In this talk, we present the design of a Cloud Multidatastore Query Language (CloudMdsQL), and its query engine. CloudMdsQL is a functional SQL-like language, capable of querying multiple heterogeneous data stores (relational and NoSQL) within a single query that may contain embedded invocations to each data store’s native query interface. The query engine has a fully distributed architecture, which provides important opportunities for optimization. The major innovation is that a CloudMdsQL query can exploit the full power of local data stores, by simply allowing some local data store native queries (e.g. a breadth-first search query against a graph database) to be called as functions, and at the same time be optimized, e.g. by pushing down select predicates, using bind join, performing join ordering, or planning intermediate data shipping. Our experimental validation, with various data stores (graph, document, relational, Spark/HDFS), and representative queries, shows that CloudMdsQL satisfies the five important requirements for a cloud multidatastore query language.

This work partially funded by the European Commission under the Integrated Project CoherentPaaS.

A new era is emerging in the world of distributed computing with the growing popularity of blockchains (shared, replicated and distributed ledgers) and the associated databases as a way of integrating inter-organizational work. Originally, the concept of a distributed ledger was invented as the underlying technology of the cryptocurrency Bitcoin. But the adoption and further adaptation of it for use in the commercial or permissioned environments is what is of utmost interest to me and hence will be the focus of this keynote. Computer companies like IBM and Microsoft, and many key players in different vertical industry segments have recognized the applicability of blockchains in environments other than cryptocurrencies. IBM did some pioneering work by architecting and implementing Fabric, and then open sourcing it. Now Fabric is being enhanced via the Hyperledger Consortium as part of The Linux Foundation. A few of the other efforts include Enterprise Ethereum, R3 Corda and BigchainDB.

While there is no standard in the blockchain space currently, all the ongoing efforts involve some combination of database, transaction, encryption, consensus and other distributed systems technologies. Some of the application areas in which blockchain pilots are being carried out are: smart contracts, supply chain management, know your customer, derivatives processing and provenance management. In this talk, I will survey some of the ongoing blockchain projects with respect to their architectures in general and their approaches to some specific technical areas. I will focus on how the functionality of traditional and modern data stores are being utilized or not utilized in the different blockchain projects. I will also distinguish how traditional distributed database management systems have handled replication and how blockchain systems do it. Since most of the blockchain efforts are still in a nascent state, the time is right for database and other distributed systems researchers and practitioners to get more deeply involved to focus on the numerous open problems.

This talk was delivered as the opening keynote at the 37th IEEE International Conference on Distributed Computing Systems (ICDCS) in Atlanta (USA) on 6 June 2017. Extensive related blockchain collateral could be found at http://bit.ly/CMbcDB

Dr. C. Mohan has been an IBM researcher for 35 years in the database area, impacting numerous IBM and non-IBM products, the research and academic communities, and standards, especially with his invention of the ARIES family of database locking and recovery algorithms, and the Presumed Abort commit protocol. This IBM (1997), and ACM/IEEE (2002) Fellow has also served as the IBM India Chief Scientist for 3 years (2006-2009).

In addition to receiving the ACM SIGMOD Innovations Award (1996), the VLDB 10 Year Best Paper Award (1999) and numerous IBM awards, Mohan was elected to the US and Indian National Academies of Engineering (2009), and was named an IBM Master Inventor (1997). This Distinguished Alumnus of IIT Madras (1977) received his PhD at the University of Texas at Austin (1981). He is an inventor of 50 patents. He is currently focused on Blockchain, Big Data and HTAP technologies (http://bit.ly/CMbcDB, http://bit.ly/CMgMDS). Since 2016, he has been a Distinguished Visiting Professor of China’s prestigious Tsinghua University. He has served on the advisory board of IEEE Spectrum, and on numerous conference and journal boards. Mohan is a frequent speaker in North America, Europe and India, and has given talks in 40 countries. He is very active on social media and has a huge network of followers.

More information could be found in the Wikipedia page at http://bit.ly/CMwIkP

Hans-Peter Kriegel has been a Professor of Informatics at Ludwig-Maximilians-Universität München, Germany, since 1991. As of April 2014 he is a Professor Emeritus due to mandatory retirement in Germany, but still has a contract with the university for research. He has published over a wide range of data mining topics including clustering, outlier detection and high-dimensional data analysis.

From the beginning of his research he was and still is working on spatial data management and similarity search, now in particular on searching and mining uncertain data, including uncertain spatio-temporal data. In 2009 the Association for Computing Machinery (ACM) elected Professor Kriegel an ACM Fellow for his contributions to knowledge discovery and data mining, similarity search, spatial data management, and access-methods for high-dimensional data. So far, his more than 450 publications have been cited approximately 50,000 times according to Google Scholar. He received both international Research Awards in Data Mining and Knowledge Discovery: the 2013 IEEE ICDM Research Contributions Award and the 2015 ACM SIGKDD Innovation Award.